Set ``litellm.preprocessing.duration_ms`` (proxy-receive -> first provider handoff) on the proxy SERVER span. ``litellm_received_at``

Persist an OAuth2 access token for a user+server pair. The payload is JSON-serialised and stored encrypted in the same ``credential_b64`` col

Transform request to Bedrock CountTokens format. Supports both Converse and InvokeModel input types. Input (Anthropic format

Filter users based on partial match of user_id or email with pagination. Behaviour depends on the ``scope_user_search_to_org`` UI-setting fl

Inline ``$ref``s backed by draft-04 ``definitions`` / OpenAPI ``components.schemas``. ``$defs`` is left untouched. Anthropic and Fireworks to

Update a new mcp server record in the db

Updates messages with model file ids. For managed files (unified file IDs), uses model_file_id_mapping if it resolves the id, otherwise

Check if the model uses custom pricing Returns True if any of `SPECIAL_MODEL_INFO_PARAMS` are present in `litellm_params` or `model_info`

Lightweight endpoint to get user info. Returns only the user object — no keys, no teams objects. This is the v2 replacement for /user/info,

OPTIONAL Validate the environment for the request

Iterate through db models for any not in router - add them. Return - number of deployments added

Add mcp_list_tools from _hidden_params to the first chunk's delta.provider_specific_fields. This method checks if MCP metadata with

Wrap a Responses-API streaming iterator so MidStreamFallbackError triggers the Router's fallback chain (parity with _acomplet

The endpoint must not return a role-block error. Detects both the 400 ``not_allowed_access`` pattern (used by most management endpoints) and

Populate ``litellm_remaining_tokens_metric`` / ``litellm_remaining_requests_metric`` from the router's internal usage counter

Attempt to repair truncated JSON produced by LLM tool calls. Handles the most common truncation patterns where the model generates valid

Realistic Bedrock ApplyGuardrail response when a topic policy fires. Mirrors the shape in ``litellm/types/proxy/guardrails/guardrail_hooks/ b

Build semantic router with tools (MCPTool objects or OpenAI function dicts).

Call anthropic_messages_handler with an Anthropic model and capture the anthropic_messages_optional_request_params dict passed to base_ll

Check that project limits respect its parent Team's limits. Mirrors _check_org_team_limits() from team_endpoints.py. Validates: - P

OpenAI tool invokes: { "role": "assistant", "content": null, "tool_calls": [ { "id": "call_abc123",

Token-count a single batch input entry's body (chat / text / embedding).

Create a custom httpx client factory that uses LiteLLM's SSL configuration. This factory follows the same CA bundle path logic as htt

Create a mock logging object

Helper to create a mock OAuth2 MCPServer.

(Helper function of add deployment) -> combined to reduce prisma db calls - Create all up list of model id's (db + config) -

Encode metadata with the upstream ephemeral key so /realtime/calls can route without requiring model as a query param.

Encrypts a dictionary of environment variables and returns them.

Convert raw OpenAI-format messages to Lasso API format with content blocks. - assistant messages with `tool_calls` → assistant messa

Return the plain-text representation of a message content value.

Fetch tools from MCP client with timeout and error handling. Uses anyio.fail_after() instead of asyncio.wait_for() to avoid conflict

Filter models by team ID. Returns models where: - team_id matches the model's BYOK team_id, OR - team_id is in access_via_team_ids, OR

Get the filtered MCP servers from the MCP server names. Fails closed when ``mcp_servers`` is explicitly provided (path- or h

Get a boto3 client for the specified service using BaseAWSLLM auth.

Get redirect parameters for Generic SSO with proper state priority handling. Optionally generates PKCE parameters if GENERIC_CLIENT_U

Get the appropriate log exporter based on the configuration.

Parse server-specific MCP auth headers from the request headers. Looks for headers in the format: x-mcp-{server_alias}-{header_name}

Get the MCP Server from the tool name (handles both prefixed and non-prefixed names) Args: tool_name: Tool name (can be

Helper for 'get_model_info'. Separated out to avoid infinite loop caused by returning 'supported_openai_param's

Determine the public model name from patch or existing model. The top-level ``model_name`` is the rename channel. For team-scoped rows the DB

Fetch all deployments for a given team_id from the database. Centralizes team deployment queries to ensure consistent filtering and error ha

Handle a PipelineExecutionResult — allow, block, or modify_response. Returns data dict if allowed, raises on block/modify_response.

Check if the model is Gemini 3 Pro or newer. Gemini 3 models include: - gemini-3-pro-preview - gemini-3-flash

Returns True if the key's `access_group_ids` expand to models that grant access to `model`. Used to let a key's access group override a team'

Helper function to list keys Args: page: int size: int user_id: Optional[str] team_id: Optional[str]

Create a mock batch database record.

Build a minimal stand-in for PrismaClient that exposes only the surface `call_with_db_reconnect_retry` actually pokes at.

Return a mock LiteLLMLoggingObj pre-configured for prompt management.

Build a mock ResponseOutputMessage with output_text parts.

Create a base64-encoded unified batch ID.

Build a valid base64-encoded unified vector-store ID.